def prepareCorpus(self, lines):
self.corpus = [[] for _ in lines]
self.words, self.wordLanguages = [], []
wIds = [{} for _ in range(self.numLang)]
numLangFlag = 0
for file, fileOff in zip(self.files, self.offsets):
for lineId, offsetId in enumerate(lines):
file.seek(fileOff[offsetId])
line = self.corpus[lineId]
for i, sentence in enumerate(file.readline().split('\t')):
languageId = i + numLangFlag
wordIds = wIds[languageId]
for word in sentence.split():
wordId = wordIds.get(word)
if wordId is None:
wordId = len(self.words)
wordIds[word] = wordId
self.words.append(word)
self.wordLanguages.append(languageId)
line.append(wordId)
numLangFlag = languageId + 1
self.wFr = [0] * len(self.words)
for line in self.corpus:
for wordId in set(line):
self.wFr[wordId] += 1
self.words.append(self.dlmn)
self.wordLanguages.append(self.numLang)
self.wFr.append(max(self.wFr) + 1)
srtfrqw = sorted(range(len(self.words)), key=self.wFr.__getitem__, reverse=True)
self.words = [self.words[i] for i in srtfrqw]
self.wordLanguages = getBestArray([self.wordLanguages[i] for i in srtfrqw], self.numLang)
new_Pos = [None] * len(self.words)
for i, wordId in enumerate(srtfrqw):
new_Pos[wordId] = i
for i, line in enumerate(self.corpus): # Replace word ids in corpus
self.corpus[i] = [new_Pos[wordId] for wordId in line]
self.wFr.sort(reverse=True)
self.wFr = getBestArray(self.wFr)
ngramgrw = range(2, self.indexer + 1)
languagegrw = range(self.numLang)
allNgramIds = [{} for _ in ngramgrw]
self.allNgrams = [[] for _ in ngramgrw]
self.ngramCorpora = [[] for _ in ngramgrw]
for line in self.corpus:
sentences = [[] for _ in languagegrw]
ngramSentences = [set() for _ in ngramgrw]
for word in line:
sentences[self.wordLanguages[word]].append(word)
for s in sentences:
s = tuple(s)
lastIdx = len(s) + 1
for one in range(2, min(self.indexer+1, lastIdx)):
ngids = allNgramIds[one-2]
ngs = self.allNgrams[one-2]
ngsen = ngramSentences[one-2]
for i in range(lastIdx - one):
ng = s[i:i+one]
ngid = ngids.get(ng)
if ngid is None:
ngid = len(ngs)
ngids[ng] = ngid
ngs.append(ng)
ngsen.add(ngid)
for one in ngramgrw:
self.ngramCorpora[one-2].append(sorted(ngramSentences[one-2]))
def __init__(self, inputFilenames):
self.maxNbLines =0
self.time = 100
self.archivos = inputFilenames
self.numNewAligns = -1
self.discontiguousFields =''
self.minSize =1
self.maxSize=7
self.dlmn=None
self.indexer =1
self.writer = HTMLOutput(sys.stdout, 'utf-8', None)
self.counter = {}
self.numAligns = 0
self.AlignedFile = getTempFIle(".al_lw")
self.offsets = []
numLines = None
self.numLang = 0
try:
self.files =[openFile(file) for file in inputFilenames]
for file in self.files:
offset = 0
fileOffsets = []
fileLanguages = None
lineNumber = -1
for lineNumber, line in enumerate(file):
fileLine = line.count('\t') + 1
if fileLanguages is None:
fileLanguages = fileLine
self.numLang += fileLine
else:
assert fileLine == fileLanguages, "There is %i columns " \
" instead of %i at line %i in file %s" % \
(fileLine, fileLanguages, lineNumber + 1, file.name)
fileOffsets.append(offset)
offset += len(line)
if numLines is None:
numLines = lineNumber + 1
else:
assert numLines == lineNumber + 1, \
"Input files have different number of lines"
self.offsets.append(getBestArray(fileOffsets))
del fileOffsets
self.minLanguages = self.numLang
narft = changeFields(self.discontiguousFields, self.numLang)
self.contiguousFields = [(i + 1 not in narft) for i in range(self.numLang)]
if self.maxNbLines < 1:
numCorpus = 1
else:
numCorpus = int(math.ceil(1. * numLines / self.maxNbLines))
self.time /= 1. * numCorpus
lines = range(numLines)
random.shuffle(lines)
for numCurpTres in range(numCorpus, 0, -1):
select = [lines.pop() for _ in range(int(math.ceil(1. * len(lines) / numCurpTres)))]
select.sort()
self.prepareCorpus(select)
self.run(self.time, self.numNewAligns)
setProbability(self.AlignedFile, self.counter, self.writer)
finally:
self.AlignedFile.close()
for file in self.files:
file.close()
